Open input does not cue the range of capability

What it is

Export PDF names itself: what it does, what kind of result you get, written on the control. A field that says “ask me anything” names nothing. Unsignaled capability range means an open text entry shows “you may type here,” not the set of operations the system actually covers. The user sees a language channel, not a map of competence.

This is not the claim that the first sentence can only be a guess — that is a first-move problem. The claim here is that the entry itself supplies no range signal: even after someone has decided to speak, they still cannot read from the surface which intents land inside the system and which hit a wall.

Why it happens

Gibson’s affordances are perceivable possibilities for action. Buttons, menus, and icons make those possibilities into objects. An empty field collapses them into a linguistic competence — “if you can write, you can use this.” Instant messaging taught the affordance “the other side understands human language,” and people migrate that look to a generative entry. The far side of a message box is a person who can refuse, follow up, or admit ignorance. The far side of a model accepts the request and fills it with fluent text. The entry looks as if anything can be said; the system covers only the patch cut by training and tool wiring.

The gulf of execution is widened by the entry itself: the goal sits in the user’s head, the legal operators are not in view. People fill the gap with world knowledge, and world knowledge is organized around conversation, not around which tools this product actually wired up.

Studying it

First use, no tutorial. Before anyone types, ask “what do you think it can do,” and code answers against the real inventory as coverage versus invention. Independent variables: whether placeholder copy names a domain, whether capability clusters sit beside the field, whether tool icons appear. Dependent variables: share of real capabilities named, count of invented ones, whether the first request lands inside competence.

Do not open with “this assistant can write code and search the web” — that spends the independent variable. A cleaner setup nests the entry in an existing product sidebar and lets people guess the range from a daily task.

Where it stops holding

When the container already narrows the range — a contract-clause assistant, a “edit this column only” pane — the signal comes from the frame, not the field, and this claim weakens. A voice assistant bound to living-room lights and locks has the room as range. People who arrived from a feature announcement already carry one slice of expectation; the empty box is no longer the only cue. This entry does not treat “too many functions to list” — that is an enumerability failure — nor whether users will blame a miss on their own wording.

Applying it

• Put a short set of verb-led capability labels in the field’s visible neighborhood, not a line that says ask anything. Labels must be things the system can actually do; fewer and true beats many and fake.
• Let placeholder copy name a domain or object (“describe the table you want changed”). Do not stretch the range to infinity with “ask me any question.”
• Make labels clickable: a click should expose the legal objects of that capability, not dump a canned sentence.
• Check: hide the product name and navigation, leave the field, ask someone new to the product what it can do. If answers drift toward omniscient Q&A and you only ship a handful of task types, the range is still unsignaled.

Related

• Same group: L2.01.2 Not knowing how to say it is the main barrier · L2.01.3 Differences in wording produce differences in results · L2.01.4 An empty box conveys no boundary; the first sentence is a guess · L2.01.5 Open input steers failure attribution toward “I said it badly” · L2.01.6 Synonymous phrasings yield different results, so users invent phrases to memorize · L2.01.7 Openness makes functions non-enumerable; the product can no longer show a complete capability list · L2.01.8 Error messages for open input stay vague because the system does not know what the user meant to do
• Nearby: L2.02 Discoverability of What Can Be Said · L2.15 Ambiguity and Clarifying Follow-ups · L1.02 Expressing Capability Boundaries
• Search terms: unsignaled capability range · gulf of execution · natural language affordance